Project 42Start learning
Module 5 of 5 · 70 min

Defend an Agentic-AI Classification

Produce a human-reviewed evidence dossier that classifies an exact system without hiding uncertainty or overstating autonomy.

Core concept

By the end

You will be able to

  • Classify a version-scoped experience across all eight taxonomy classes.
  • Support every material claim with documented, tested, inferred, or unknown evidence.
  • Compare deterministic, single-agent, and multi-agent alternatives using equivalent cases.
  • Earn mastery through knowledge score, critical gates, artifacts, and human review.
01

Freeze the unit of classification

Record the product surface, versions, configuration, model, tools, permissions, state services, runtime, observation date, and accountable owner. Exclude behavior you did not observe or cannot source.

State the question precisely. A useful dossier answers what this configured experience did and who controlled it, not whether an entire brand is intelligent, autonomous, or agentic.

02

Build the claim and behavior ledger

For each classification signal, attach a first-party source or retained run artifact. Keep documented and tested evidence separate, identify inferences, and preserve unknowns with a resolution plan.

Trace next-step selection, tools, external effects, state, feedback, coordination, stop conditions, recovery, and human authority. A missing critical boundary prevents a confident agentic classification.

03

Compare the least complex alternatives

Use equivalent cases to compare a deterministic workflow, bounded single-agent design, and multi-agent design where applicable. Measure outcome quality, safety, latency, cost, coordination overhead, recovery, trajectory, and review load.

Do not reward autonomy or agent count. Select the least complex design that passes critical gates and meets the declared need.

04

Make uncertainty and authority visible

The final disposition states the classification, supporting evidence, rejected alternatives, residual unknowns, expiry date, and triggers for re-review.

A qualified person reviews the evidence and critical errors. The system or model being classified cannot approve its own mastery or publication.

Practice activity

Create and defend the classification dossier

  1. Freeze one exact experience and create its layer, control-flow, and responsibility maps.
  2. Build a dated claim ledger with documented, tested, inference, and unknown labels.
  3. Compare equivalent deterministic, single-agent, and multi-agent designs when the evidence supports them.
  4. Submit the classification, residual uncertainty, review date, and human disposition for challenge and revision.

What to produce

  • Four required dossier artifacts with traceable claims and no critical classification errors.
  • A recorded human review that accepts, rejects, or returns the dossier with corrective evidence.

Reflect before continuing

What evidence would most likely cause you to change this classification?

Applied capstone

Agentic-AI Classification Dossier

Pass at 80%

Classify one exact product experience and defend the result with current sources, run evidence, comparative design evidence, unknowns, and human review.

Required artifacts

Provide a path, URL, or short stable reference for every artifact.

Evidence rubric

Award whole points from 0 to the criterion maximum. The total is checked against the published rubric.

Exact scope and taxonomy

The dossier names the exact experience and applies the eight-class taxonomy without provider generalization or marketing-label shortcuts.

Evidence required
  • Experience scope
  • Layer map
  • Classification decision
Map this score to evidence

Choose one or more submitted artifacts or a recorded knowledge check.

Control and responsibility boundaries

Next-step control, tools, effects, state, feedback, coordination, stopping, recovery, and human authority are accurately located.

Evidence required
  • Control-flow trace
  • Responsibility map
  • Terminal outcomes
Map this score to evidence

Choose one or more submitted artifacts or a recorded knowledge check.

Evidence quality and freshness

Claims use current primary sources and retained tests while documentation, testing, inference, and unknowns remain distinct.

Evidence required
  • Claim ledger
  • Source dates
  • Run evidence
  • Unknown resolution plan
Map this score to evidence

Choose one or more submitted artifacts or a recorded knowledge check.

Equivalent design comparison

The learner compares appropriate deterministic, single-agent, and multi-agent alternatives on equivalent cases and rejects unnecessary complexity.

Evidence required
  • Equivalent cases
  • Measured results
  • Complexity decision
Map this score to evidence

Choose one or more submitted artifacts or a recorded knowledge check.

Critical classification safety

The dossier contains no tool-call-equals-agent, model-harness collapse, hidden unknown, missing authority, or unsupported autonomy error.

Evidence required
  • Critical-error checklist
  • Corrective revision evidence
Map this score to evidence

Choose one or more submitted artifacts or a recorded knowledge check.

Human disposition

An accountable reviewer records the decision, residual uncertainty, expiry, and re-review triggers without delegating approval to a model.

Evidence required
  • Reviewer disposition
  • Residual uncertainty
  • Review deadline
Map this score to evidence

Choose one or more submitted artifacts or a recorded knowledge check.

Evidence

Sources and verification

Knowledge check

Make it stick.

Pass at 80%

Choose the strongest answer for each question. Your attempts become part of your device-local transcript.

01What must a classification dossier freeze first?
02Which evidence combination supports an agentic-system classification?
03When should a multi-agent design be selected?
04Can a score above 80 percent override a critical classification error?
05Who approves the final applied mastery disposition?